AutoFox
8-second pages → under 2 seconds
Rescued a slow Bubble app for dealerships — dropped page loads from 8 seconds to under 2 and stabilised the AI image pipeline.
- 0%
- faster pages
- 0%
- faster turnover
What was blocking progress
- App painfully slow — 8-second page loads
- AI integration kept breaking
- No proper error handling
The situation
AutoFox is an AI-powered vehicle imagery platform for dealerships. The owner had built it on Bubble but the app was painfully slow — 8-second page loads — and the AI integration kept breaking with no error handling.
The Bubble database had grown without optimisation. Every page loaded full lists of vehicles and images without constraints. The API Connector calls to Replicate for AI image processing had no timeout handling, no retry logic and no user feedback when processing failed.
The dealership teams using the app daily were losing patience. The performance problems weren't a Bubble limitation — they were implementation problems that a Bubble developer who understood data architecture and API workflows could fix.
What we built
We audited the Bubble app's performance, restructured the database and rebuilt the API integration with proper error handling.
Database performance overhaul
The Bubble database had vehicle records, images and processing jobs stored in data types with no indexing strategy. Every page ran unfiltered "Do a search for" queries that loaded thousands of records. We restructured with tight search constraints, pagination and deferred loading using Bubble's "Only when" conditions. Page loads dropped from 8 seconds to under 2.
API Connector reliability
The Replicate AI integration was built as direct API calls with no error handling. We rebuilt it using Bubble API workflows with proper timeout handling, retry logic and status tracking. Each image processing job now has a status field (queued, processing, complete, failed) that the frontend polls using scheduled workflows.
Batch processing workflow
Dealerships need to process multiple vehicle photos in a single session. We built a batch processing flow in Bubble where dealers upload a set of photos, the backend queues them through scheduled API workflows, and results appear as each image completes. Custom states on the processing page show real-time progress without page refreshes.
Image storage optimisation
Original images and AI-processed outputs were stored without compression or size management. We added image processing via Imgix and proper file management in Bubble's storage. Thumbnail previews load first, with full-resolution images loading on demand.
- Database restructure with constrained searches and deferred loading
- Reliable API Connector integration with error handling and retry logic
- Batch processing workflow with real-time progress tracking
- Image storage optimisation with Imgix integration
Technical challenge
The AI image pipeline had to handle large vehicle photos, process them through Replicate models for background removal and enhancement, and return production-quality results within seconds.
The result
Page load times dropped from 8 seconds to under 2. Dealers produce production-ready vehicle visuals in under 5 minutes per vehicle. The Bubble app now handles daily use by dealership teams without performance complaints.
The AI image pipeline processes photos reliably with proper error handling. When Replicate returns an error, the Bubble app retries automatically and notifies the user only if the retry fails. 25% improvement in inventory turnover reported by the client.
Have a Bubble app that's too slow for real users?
Next case study
Huntr AI-powered SaaS